You don't say whether the missing values are in the dependent variable or the independent variables. If in the dependent variable, I would say to delete those observations from the fitting part (you could use them to predict values and check the appropriateness of your model). For the independent values, you might look at PROC EXPAND, which would enable interpolation of missing values.
Also, how many records do you have? Since you are looking at 54 independent variables, you probably need around 300 complete observations to get stable results.
Since VARMAX does not support missing values, interpolation or multiple imputation are the only alternatives I see for a vector autoregressive model using VARMAX. You might look at some of the state space examples to see if they can provide you with a road forward.
Also, try asking this over in the Forecasting & Econometrics community. Perhaps someone with more experience in time series analysis could give you some help.
SteveDenham
... View more